The mold change does not start without visual evidence

The problem

Two chained costs, and only one of them is visible.

The mold change has two chained costs and only one of them is visible. The visible one is the start-up queue: line hours and kilos of resin going to regrind until weight and wall thickness distribution stabilise. It is accepted as inevitable and is not measured comparably between shifts or between lines. The expensive one appears now and then: if the purge was not complete — and the risk rises when recycled content or color changes — the first batch of the new reference comes out carrying the previous material and has to be held in full. That batch has already consumed resin, machine hours and labour. Behind both sits the same cause: "ready to start" depends on somebody saying so, with no photo, no objective comparison and no record. When something goes wrong there is no way to reconstruct what state the machine was actually in at start-up.

• The visible one is the start-up tail: line hours and kilos of resin going to regrind until weight and thickness distribution stabilize. It is assumed as unavoidable and is not measured comparably between shifts or lines.
• The expensive one shows up occasionally: if purging was not complete, the first lot of the new part number comes out with the previous material and has to be blocked entirely.
• The risk of incomplete purging rises exactly when recycled content or color changes, which is increasingly often.
• And "ready to start" is today a signature with no objective evidence behind it.

How it fits the IRIS system

Edge with JIDOKA AI and SMED AI — one brakes, the other shortens.

Edge + JIDOKA AI + SMED AI. Fixed cameras on the points that decide the start-up. Step by step:

The machine does not start because the system says so: it starts when the evidence supports it and a person signs on that evidence. That distinction matters for anyone worried a camera will stop their line.

• Reference state. For each reference, the "ready to start" state is defined: correct and complete mold, cavities clear, purge finished, hopper in condition.
• Visual comparison. Edge cameras on the mold carrier, the blow nozzle, the outfeed conveyor and the feed hopper compare the current state against that reference. Each point issues its own verdict.
• Weight control. The first parts are weighed and checked against the target window.
• JIDOKA AI block. The line does not start until every point confirms. The machine does not start because the system says so: it starts when the manager signs against that evidence. The signature stays human, but now it is backed up.
• SMED AI. The system guides the changeover sequence and times each step, so the change stops being a ritual of variable duration and becomes a measured process, comparable between shifts and between lines.
• Automatic evidence. The before and after photos of each point feed the audit evidence pack directly.

See the full IRIS architecture →

Before and after

Today's mold change versus the validated change

Aspect	Today	With iLEAN Edge
Start-up validation	A blind signature	Signature on point-by-point visual evidence
Kilos lost per changeover	Whatever they turn out to be	Estimated ≥30 % reduction
Batch contaminated by incomplete purge	Happens occasionally	Eliminated
Mold and cavities	Checked by eye	Against the reference state for that part number
Weight of the first pieces	Groped for	Checked against the target window
Changeover duration	Not comparable between shifts	Measured and guided by SMED AI

Impact estimate

Estimated impact — to validate with your own numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• Estimated payback 5-10 months in a plant with high mix and frequent mold changes.
• The return is measured in OEE points recovered and in kilos of resin that stop going to regrind, with an estimated ≥30 % reduction.
• Batch contaminated by incomplete purging: eliminated, and it is the cost nobody budgets for because it only appears occasionally.
• Changeover duration measured comparably between shifts and lines, which it is not today.

Estimated payback 5-10 months in a plant with a high mix and frequent mold changes. The return is measured directly in OEE points recovered and in kilos of resin that stop going to regrind, with an estimated reduction of at least 30% in kilos lost per changeover. On top of that sits the elimination of the batch held for an incomplete purge, an infrequent but very expensive event. *Estimate to be validated.*

And the fair question from the production manager

«What if the camera blocks my line for no reason?» — conformity is checked against the reference state defined for that specific part number, not against a generic criterion. In case of doubt the system does not block silently: it escalates to the supervisor, who signs on the evidence. It stops on a verifiable discrepancy, not on model uncertainty.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.